DsgeSmoother

 stephane.adjemian@cepremap.cnrs.fr [09-07-2004]

 Adapted from mj_optmumlik.m

0001 function [alphahat,etahat,epsilonhat,ahat,SteadyState,trend_coeff,aK] = DsgeSmoother(xparam1,gend,Y)
0002 % stephane.adjemian@cepremap.cnrs.fr [09-07-2004]
0003 %
0004 % Adapted from mj_optmumlik.m
0005   global bayestopt_ M_ oo_ estim_params_ options_
0006 
0007   alphahat     = [];
0008   epsilonhat    = [];
0009   etahat        = [];
0010   nobs         = size(options_.varobs,1);
0011   smpl        = size(Y,2);
0012 
0013   set_all_parameters(xparam1);
0014 
0015   %------------------------------------------------------------------------------
0016   % 2. call model setup & reduction program
0017   %------------------------------------------------------------------------------
0018   [T,R,SteadyState] = dynare_resolve;
0019   bayestopt_.mf = bayestopt_.mf2;
0020   if options_.loglinear == 1
0021     constant = log(SteadyState(bayestopt_.mfys));
0022   else
0023     constant = SteadyState(bayestopt_.mfys);
0024   end
0025   trend_coeff = zeros(nobs,1);
0026   if bayestopt_.with_trend == 1
0027     trend_coeff = zeros(nobs,1);
0028     nx1 = estim_params_.nvx+estim_params_.nvn+estim_params_.ncx+estim_params_.ncn;
0029     for i=1:nobs
0030       trend_coeff(i) = evalin('base',bayestopt_.trend_coeff{i});
0031     end
0032     trend = constant*ones(1,gend)+trend_coeff*(1:gend);
0033   else
0034     trend = constant*ones(1,gend);
0035   end
0036   start = options_.presample+1;
0037   np    = size(T,1);
0038   mf    = bayestopt_.mf;
0039   % ------------------------------------------------------------------------------
0040   %  3. Initial condition of the Kalman filter
0041   % ------------------------------------------------------------------------------
0042   %
0043   %  C'est ici qu'il faut dterminer Pinf et Pstar. Si le modle est stationnaire,
0044   %  alors il suffit de poser Pstar comme la solution de l'uation de Lyapounov et
0045   %  Pinf=[].
0046   %
0047   Q = M_.Sigma_e;
0048   H = M_.H;
0049   
0050   if options_.lik_init == 1        % Kalman filter
0051     Pstar = lyapunov_symm(T,R*Q*transpose(R));
0052     Pinf    = [];
0053   elseif options_.lik_init == 2 % Old Diffuse Kalman filter
0054     Pstar = 10*eye(np);
0055     Pinf    = [];
0056   elseif options_.lik_init == 3 % Diffuse Kalman filter
0057     Pstar = zeros(np,np);
0058     ivs = bayestopt_.i_T_var_stable;
0059     Pstar(ivs,ivs) = lyapunov_symm(T(ivs,ivs),R(ivs,:)*Q* ...
0060                    transpose(R(ivs,:)));
0061     Pinf  = bayestopt_.Pinf;
0062     % by M. Ratto
0063     RR=T(:,find(~ismember([1:np],ivs)));
0064     i=find(abs(RR)>1.e-10);
0065     R0=zeros(size(RR));
0066     R0(i)=sign(RR(i));
0067     Pinf=R0*R0';
0068     % by M. Ratto
0069   end
0070   % -----------------------------------------------------------------------------
0071   %  4. Kalman smoother
0072   % -----------------------------------------------------------------------------
0073   if estim_params_.nvn
0074     if options_.kalman_algo == 1
0075       [alphahat,epsilonhat,etahat,ahat,aK] = DiffuseKalmanSmootherH1(T,R,Q,H,Pinf,Pstar,Y,trend,nobs,np,smpl,mf);
0076       if all(alphahat(:)==0)
0077     [alphahat,epsilonhat,etahat,ahat,aK] = DiffuseKalmanSmootherH3(T,R,Q,H,Pinf,Pstar,Y,trend,nobs,np,smpl,mf);
0078       end
0079     elseif options_.kalman_algo == 3
0080       [alphahat,epsilonhat,etahat,ahat,aK] = DiffuseKalmanSmootherH3(T,R,Q,H,Pinf,Pstar,Y,trend,nobs,np,smpl,mf);
0081     end
0082   else
0083     if options_.kalman_algo == 1
0084       [alphahat,etahat,ahat,aK] = DiffuseKalmanSmoother1(T,R,Q,Pinf,Pstar,Y,trend,nobs,np,smpl,mf);
0085       if all(alphahat(:)==0)
0086     [alphahat,etahat,ahat,aK] = DiffuseKalmanSmoother3(T,R,Q,Pinf,Pstar,Y,trend,nobs,np,smpl,mf);
0087       end
0088     elseif options_.kalman_algo == 3
0089       [alphahat,etahat,ahat,aK] = DiffuseKalmanSmoother3(T,R,Q,Pinf,Pstar,Y,trend,nobs,np,smpl,mf);
0090     end
0091   end